Cross-region modeling joint processing method and system based on mask constraint and multi-modal adaptive attenuation fusion

Abstract

The present application relates to the technical field of geographic information processing, and provides a cross-region modeling edge processing method and system based on mask constraint and multi-modal adaptive attenuation fusion. The method comprises the following steps: step 1: obtaining grid data generated by independent modeling of multiple target regions, unifying the coordinate systems of the grid data and completing spatial splicing to obtain an initial edge result; step 2: extracting a shared boundary between adjacent target regions from the initial edge result, and performing spatial sampling on both sides of the shared boundary to calculate regional pixel values; step 3: performing level mapping on the regional pixel values according to a preset hierarchical system, identifying edge abnormal points of the initial edge result according to the level mapping result, performing spatial clustering on the edge abnormal points, and generating continuous abnormal line segments based on the spatial clustering result; and step 4: using the continuous abnormal line segments to perform spatial adaptive fusion repair on the initial edge result to obtain target grid data.

Description

Technical Field

[0001] This invention relates to the field of geographic information processing technology, and in particular to a cross-regional modeling edge processing method and system based on mask constraint and multimodal adaptive attenuation fusion. Background Technology

[0002] In spatial information processing and large-scale surface parameter inversion modeling, due to the wide research scope and significant surface heterogeneity, it is usually difficult to use a single global model for high-precision fitting. Existing technologies usually adopt the method of dividing administrative divisions or natural geographical divisions to divide a large area into several sub-regions. Then, each sub-region is independently modeled and generates local spatial raster data. Finally, the entire area is stitched together to form a complete spatial data product.

[0003] However, due to inherent differences in basic sample distribution, characteristic variable sets, modeling algorithm structure, and parameter tuning among different administrative or natural regions, spatial discontinuities are easily generated at shared regional boundaries when the independent model results of each sub-region are stitched together globally. These discontinuities manifest as abrupt numerical changes, jumps in discrete levels, and artificial faults (i.e., striping effects or edge faults) that appear visually and statistically distinct in a banded distribution. This artificial model error introduced by stitching not only disrupts the spatial autocorrelation of the global data but also reduces the reliability of subsequent global spatial analysis applications.

[0004] For handling splicing seams, existing technologies typically employ mean smoothing with a fixed buffer, global linear transition methods, or simple hard-cut splicing. However, these methods have the following significant limitations: they cannot accurately locate and effectively distinguish between artificial model faults and real-world natural geographic features; they lack topologically coherent geometric representations of discrete anomalies; and excessive smoothing can lead to the loss of the spatial structure and high-frequency texture features of the original raster. Summary of the Invention

[0005] To address the issues of discontinuous seams and loss of spatial details during the stitching process of existing multi-source spatial attribute raster data, this invention proposes a cross-regional modeling edge processing method and system based on mask constraints and multimodal adaptive attenuation fusion. By constructing a closed-loop technical system of dynamic mask pre-repair, quantitative detection of feature normals, multimodal adaptive processing, and full-modal result comparison, it achieves accurate interception of edge spatial anomalies, adaptive seamless repair, and intelligent selection of optimal modes.

[0006] In a first aspect, the present invention provides a cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion, comprising:

[0007] Step 1: Obtain raster data generated independently from multiple target regions, unify the coordinate system of the raster data and complete spatial stitching to obtain the initial edge-joining result;

[0008] Step 2: Extract the shared boundary between adjacent target regions from the initial edge-matching results, and perform spatial sampling on both sides of the shared boundary to calculate the region cell value;

[0009] Step 3: Map the pixel values ​​of the region according to the preset hierarchical system, identify the edge anomalies of the initial edge connection result according to the hierarchical mapping result, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering result.

[0010] Step 4: Using the continuous abnormal line segments, perform spatial adaptive fusion repair on the initial edge-joining results to obtain the target raster data.

[0011] Furthermore, step 1 specifically includes:

[0012] Acquire raster data and corresponding administrative division vector maps generated independently from multiple target areas, and unify the raster data of multiple target areas to the reference coordinate system through reprojection technology;

[0013] The administrative vector boundaries of multiple target regions passing through coordinate system one are extracted, and the border area is determined outside the administrative vector boundary according to a preset distance. The border area of ​​multiple target regions is fused pixel by pixel using an overlapping filtering mean strategy to obtain the initial border result.

[0014] Furthermore, step 1 specifically includes:

[0015] Acquire raster data and corresponding administrative division vector maps generated independently from multiple target areas, and unify the raster data of multiple target areas to the reference coordinate system through reprojection technology;

[0016] Extract the administrative vector boundaries of multiple target areas passing through coordinate system one, and crop and stitch the raster data of the target areas according to the administrative vector boundaries;

[0017] The administrative division vector regions corresponding to the stitched raster data are converted into mask control matrices.

[0018] Using the mask control matrix as a spatial constraint, an interpolation method based on inverse distance weights is employed to fill in the stitched raster data, resulting in the initial edge-joining result.

[0019] Further, in step 2, spatial sampling is performed on both sides of the shared boundary to calculate the region cell value, including:

[0020] On the shared boundary, sampling center points are extracted at preset intervals, and based on the orthogonal normal direction of the sampling center points on the shared boundary, they are extended to both sides of the region to a preset buffer distance to obtain normal sampling lines;

[0021] Multiple sampling points are extracted along the normal sampling line with a preset fixed step size. The effective pixel statistical mean of each sampling point within its local neighborhood window is calculated. The arithmetic mean of the effective pixel statistical mean of all sampling points in a single-sided region of the normal sampling line is calculated as the region pixel value.

[0022] Furthermore, step 3 specifically includes:

[0023] Step 3.1: Map the pixel values ​​of the regions on both sides of the shared boundary to the corresponding discrete levels according to the preset hierarchical system, calculate the absolute value of the difference between the discrete levels on both sides, and determine the edge connection anomalies of the initial edge connection result based on the absolute value.

[0024] Step 3.2: Construct a circular buffer with each of the aforementioned edge anomaly points as the center and a preset clustering topological distance as the radius, and obtain a spatially connected component polygon by merging the intersecting circular buffers;

[0025] Step 3.3: For the edge anomalies that fall into the same spatially connected component polygon, sort and connect them using the nearest neighbor distance algorithm to generate continuous anomaly line segments.

[0026] Furthermore, step 4 specifically includes:

[0027] Step 4.1: Smooth the initial edge-joining result to obtain smoothed base data, and construct the spatial distance field between the continuous abnormal line segments and the initial edge-joining result;

[0028] Step 4.2: Generate different spatial adaptive decay fusion weights based on the spatial distance field, and use the different spatial adaptive fusion weights to perform weighted fusion on the initial edge repair results and the smooth basis function to obtain multiple sets of edge repair data;

[0029] Step 4.3: Using the edge repair data of each group as the initial edge repair results, re-execute steps 2 and 3 respectively to generate the corresponding continuous abnormal line segments, and determine the final target raster data based on the number of continuous abnormal line segments in each group.

[0030] Furthermore, step 4.1 specifically includes:

[0031] The initial edge-joining results are globally smoothed using a Gaussian low-pass filtering algorithm to obtain smoothed base data without numerical faults.

[0032] The spatial distance field is obtained by calculating the spatial distance from each pixel in the initial edge-connection result to the continuous abnormal line segment based on the Euclidean distance transformation algorithm.

[0033] Furthermore, the different spatial adaptive decay fusion weights are calculated using the traditional Gaussian decay mode, the cubic smooth step mode, the quintic smooth step mode, and the linear decay mode, respectively; wherein the spatial adaptive decay fusion weights corresponding to the traditional Gaussian decay mode are... The calculation formula is:

[0034]

[0035] in, This represents the distance from a pixel to a continuous abnormal line segment. Indicates the attenuation control parameter. This indicates the set buffer distance;

[0036] The spatial adaptive decay fusion weights corresponding to the three smooth step modes The calculation formula is:

[0037]

[0038] In the formula,

[0039] ;

[0040] The spatial adaptive decay fusion weights corresponding to the five smooth step modes The calculation formula is:

[0041] ;

[0042] The spatial adaptive decay fusion weight corresponding to the linear decay mode The calculation formula is:

[0043] .

[0044] Furthermore, step 4.3 specifically includes:

[0045] Using the edge repair data of each group as the initial edge repair results, steps 2 and 3 are re-executed to generate corresponding continuous abnormal line segments as residual abnormal line segments. The total length of each group of residual abnormal line segments is calculated, and the formula is used to... Calculate the abnormal line segment elimination rate for each group, where This indicates the rate of elimination of abnormal line segments. This represents the total length of the continuous abnormal line segments before repair. Indicates the total length of the remaining abnormal line segments;

[0046] The edge repair data with the highest abnormal line segment elimination rate was selected as the target raster data.

[0047] Secondly, the present invention provides a cross-region modeling and edge-joining processing system based on mask constraint and multimodal adaptive attenuation fusion, comprising:

[0048] The initial edge-joining unit is used to acquire raster data generated independently by modeling multiple target areas, unify the coordinate system of the raster data and complete spatial stitching to obtain the initial edge-joining result;

[0049] A spatial sampling unit is used to extract the shared boundary between adjacent target regions from the initial edge-connection result, and to perform spatial sampling on both sides of the shared boundary to calculate the region cell value;

[0050] An anomaly identification unit is used to perform level mapping on the pixel values ​​of the region according to a preset hierarchical system, identify the edge anomalies of the initial edge connection result according to the level mapping result, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering result.

[0051] The edge repair unit is used to perform spatial adaptive fusion repair on the initial edge repair result using the continuous abnormal line segments to obtain the target raster data.

[0052] The beneficial effects of this invention are as follows:

[0053] The method provided by this invention first acquires regional pixel values ​​through normal sampling, which effectively reduces the impact of local noise and improves the stability of edge detection. Secondly, it achieves automatic identification of edge anomalies through difference judgment based on a hierarchical system. Then, through spatial clustering and structural reconstruction, discrete anomaly edge points are transformed into a spatial structure of continuous anomaly line segments, enhancing the interpretability of the anomaly region and the targeted nature of subsequent processing. Furthermore, it provides multimodal spatial adaptive attenuation fusion weights, constructing a smooth base and distance weight field in the continuous anomaly line segment region, achieving seamless transition and effectively eliminating the striping effect common in traditional stitching methods while preserving the texture features of the original data. In addition, this invention uses a closed-loop evaluation mechanism to quantify the repair effect, thereby improving the overall processing quality. In summary, the method provided by this invention has strong versatility and can be widely applied in fields such as remote sensing data fusion, environmental element analysis, and integration of multi-region prediction results. Attached Figure Description

[0054] Figure 1 A flowchart illustrating a cross-region modeling edge processing method based on mask constraint and multimodal adaptive attenuation fusion provided in an embodiment of the present invention;

[0055] Figure 2 A schematic diagram of a shared boundary provided for an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram showing the full and partial magnification of the sampling ray provided in an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of spatial neighborhood feature extraction of sampling points provided in an embodiment of the present invention;

[0058] Figure 5 This is a schematic diagram of abnormal joint points provided in an embodiment of the present invention;

[0059] Figure 6 This is a schematic diagram of continuous abnormal line segments provided in an embodiment of the present invention;

[0060] Figure 7 The diagram illustrates the target raster data and the unrepaired stitched data provided in the embodiments of the present invention; wherein (a) is a diagram of the target raster data processed by the method of the present invention, and (b) is a diagram of the unrepaired stitched data.

[0061] Figure 8 This is a structural diagram of a cross-region modeling and edge processing system based on mask constraint and multimodal adaptive attenuation fusion, provided in an embodiment of the present invention. Detailed Implementation

[0062] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of the embodiments of this invention will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0063] like Figure 1 As shown, this embodiment of the invention provides a cross-region modeling edge processing method based on mask constraint and multimodal adaptive attenuation fusion, including:

[0064] S101: Obtain raster data generated independently from multiple target areas, unify the coordinate system of the raster data and complete spatial stitching to obtain the initial edge-joining result.

[0065] Specifically, first, the data files (e.g., *.tif) of the raster datasets for each target region are read, and their coordinate reference systems are extracted. If inconsistencies exist in the coordinate systems, WarpedVRT reprojection technology is used to unify them to the reference coordinate system. Second, the raster matrices of multiple target regions are concatenated into a global initial matrix, which serves as the initial edge-joining result.

[0066] S102: Extract the shared boundary between adjacent target regions from the initial edge-connection result, and perform spatial sampling on both sides of the shared boundary to calculate the region cell value.

[0067] Specifically, the initial edge-joining result obtained by S101 has splicing marks, and the shared vector boundary of two adjacent target regions is obtained (such as...). Figure 2 As shown in the figure, the shared boundary is discretized, then spatial sampling is performed, and the region cell values ​​are calculated for the regions on both sides of the shared boundary based on the spatial sampling results.

[0068] S103: Map the regional pixel values ​​according to the preset hierarchical system, identify the edge anomalies in the initial edge connection results based on the hierarchical mapping results, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering results.

[0069] Specifically, a preset analysis system is introduced to map continuous regional pixel values ​​into discrete level variables. In this embodiment, these are mapped to normal, level 1, level 2, and severe. Level 1, level 2, and severe are judged as edge anomaly points. In order to obtain the continuous geometric shape of the abnormal seam, the edge anomaly points are reconstructed into continuous abnormal line segments based on the spatial clustering results.

[0070] S104: Using continuous abnormal line segments, spatial adaptive fusion repair is performed on the initial edge-joining results to obtain the target raster data.

[0071] Specifically, continuous abnormal line segments are used as the core area of ​​abnormal seams for abnormal repair, while the real texture features of the original modeling grid are preserved in areas far from abnormal areas, so as to achieve a seamless and natural transition in space.

[0072] The method provided in this invention identifies abnormal seam areas through spatial sampling and repairs these abnormal seam areas, thereby achieving seamless and natural multi-region modeling and improving the usable quality of the fused raster data.

[0073] As one possible implementation method, S101 performs spatial stitching using an overlapping mean transition strategy, specifically including:

[0074] The raster data and corresponding administrative division vector maps generated independently from multiple target areas are acquired, and the raster data of multiple target areas are unified to the reference coordinate system through reprojection technology.

[0075] Specifically, the raster data files and corresponding administrative division county-level Shapefile vector base maps of multiple adjacent target areas are obtained through independent modeling. Their coordinate reference systems are extracted. If the coordinate systems are inconsistent, WarpedVRT reprojection technology is used to unify them to the reference coordinate system.

[0076] The administrative vector boundaries of multiple target regions passing through coordinate system one are extracted, and the border areas are determined outside the administrative vector boundaries according to a preset distance. The overlapping filtering mean strategy is used to perform pixel-by-pixel mean fusion of the border areas of multiple target regions to obtain the initial border results.

[0077] Specifically, the administrative vector boundary (polygon) of the current county is extracted and a buffer expansion surface with a preset distance (set to 200m in this embodiment) is constructed outward to pre-clip the grid of the county. When using the grid tiling merging function merge, the overlapping area merging method is set to method='mean', that is, the arithmetic mean of each pixel is calculated for the overlapping outer strips of each county to achieve a preliminary smooth transition, and the result is used as the initial edge-joining result.

[0078] As one possible implementation method, S101 performs spatial splicing through a strict boundary shearing strategy, specifically including:

[0079] Acquire raster data and corresponding administrative division vector maps generated independently from multiple target areas, and unify the raster data of multiple target areas to the reference coordinate system through reprojection technology;

[0080] Extract the administrative vector boundaries of multiple target areas passing through coordinate system one, and then crop and stitch the raster data of the target areas according to the administrative vector boundaries;

[0081] Specifically, the original administrative boundaries of the county are strictly cut. At this point, because the rasterized rows and columns of each area cannot be perfectly aligned, small NoData gaps will appear at the strict boundary between the two areas after splicing due to geometric misalignment.

[0082] The administrative division vector regions corresponding to the stitched raster data are converted into mask control matrices.

[0083] Specifically, the process iterates through the fields (such as XZQMC) in the administrative division vector base map, extracts the polygons of counties and districts that actually have data and are participating in the evaluation, performs a geometric union operation (unary_union) to generate a vector polygon (Footprint) of the entire legal coverage area, and uses the geometry_mask module to perform matrix transformation on the Footprint. It returns a True signal (marked as a legal filling area) inside the polygon and a False signal (marked as an out-of-bounds no-data area) outside the county-level administrative division.

[0084] Using the mask control matrix as a spatial constraint, an interpolation method based on inverse distance weights is employed to fill in the stitched raster data, resulting in the initial edge-joining result.

[0085] Specifically, the `fillnodata` function is used to extract truly valid pixels from the original raster (i.e., non-empty, non-NaN pixels greater than -99999.0) as the sole interpolation emission source, extending outwards to the surrounding holes. The maximum search distance is set to 150 pixels. After the interpolation calculation is complete, conditional filtering is performed using a mask control matrix.

[0086]

[0087] Meanwhile, to ensure that the original valid cell values ​​are not accidentally tampered with, a layer of original valid cell locking matrix is ​​overlaid on the interpolation result. This mechanism strictly locks the interpolation region within the legal administrative territory and never extrapolates, resulting in pure and seamless global initial edge-connecting raster data.

[0088] As one possible implementation, in S102, spatial sampling is performed on both sides of the shared boundary to calculate the region cell value, including:

[0089] On the shared boundary, sampling center points are extracted at preset intervals, and based on the orthogonal normal direction of the sampling center points on the shared boundary, they are extended to both sides of the region to a preset buffer distance to obtain normal sampling lines.

[0090] Specifically, the shared boundary is discretized, and interpolation is performed at preset intervals (e.g., 30 meters) to generate a series of sampling center points. Each sampling center point has a globally unique identifier (CID), and the local tangent azimuth of the boundary at that point is calculated using the interpolated coordinates of its immediate and nearest neighboring minima. And then add it up. Obtain the orthogonal normal angle Starting from the current sampling center point, extend along the orthogonal normal direction to regions A and B respectively, up to the set normal sampling buffer distance limit (e.g., BUFFER_DISTANCE = 600m), reconstructing the bidirectional normal sampling rays ray_A and ray_B (e.g., ...). Figure 3 (As shown).

[0091] Multiple sampling points are extracted along the normal sampling line with a preset fixed step size. The effective pixel statistical mean of each sampling point within its local neighborhood window is calculated. The arithmetic mean of the effective pixel statistical mean of all sampling points on one side of the normal sampling line is calculated as the region pixel value.

[0092] Specifically, to eliminate the influence of local high-frequency random noise, a local spatial neighborhood window analysis method is used to extract feature values ​​along the ray (such as...). Figure 4 (As shown). The specific operation is as follows: On a single-sided sampling ray, a series of detailed sampling points are generated by interpolation from the center point outwards at a fixed step size (e.g., 30m). The geographic projection coordinates of each detailed sampling point are then calculated using the inverse affine transformation formula. Mapped to the row and column matrix indices of the initial edge-joining result raster data Extracting from the index as the center, A local pixel neighborhood window of a certain size is used. After removing NoData or NaN outliers within the window, the arithmetic mean of the valid pixels is calculated as the spatial feature value of that detailed sampling point.

[0093]

[0094] The spatial feature values ​​of all valid detailed sampling points extracted from the sampling line on side A are arithmetically averaged again to obtain the regional pixel value (A_RepVal) of the region on side A at the boundary point; similarly, the regional pixel value (B_RepVal) of the region on side B is calculated.

[0095] As one possible implementation method, S103 specifically includes:

[0096] S103.1: Map the pixel values ​​of the regions on both sides of the shared boundary to the corresponding discrete levels according to the preset hierarchical system, calculate the absolute value of the difference between the discrete levels on both sides, and determine the edge outliers in the initial edge-joining result based on the absolute value (e.g., ...). Figure 5 (As shown).

[0097] Specifically, a pre-established business objective grading standard dictionary is introduced as a preset grading system, mapping the continuous region pixel values ​​A_RepVal and B_RepVal proposed in S102 to discrete grading variables. The grading standard dictionary defines the grading boundary ranges (containing a minimum value min and a maximum value max) for each indicator (such as organic matter, total nitrogen, pH value, etc.). If a representative value exceeds the global maximum / minimum value defined in the dictionary, it is forcibly truncated and mapped to the corresponding boundary grading, ultimately transforming the continuous values ​​into discrete business grading. and Calculate the absolute difference in grade between the regions on both sides of the same boundary center point, and classify the edge consistency based on this difference: when the grade difference is less than or equal to 1, the edge position is judged to be natural and marked as "normal"; when the grade difference is equal to 2, it is judged as "Level 1 anomaly"; when the grade difference is equal to 3, it is judged as "Level 2 anomaly"; when the grade difference is greater than 3, it is judged as "severe anomaly". Boundary center points marked as Level 1, Level 2 and severe anomalies are collectively referred to as edge anomaly points.

[0098] S103.2: Construct a circular buffer with each edge anomaly point as the center and a preset clustering topological distance as the radius, and obtain a spatially connected component polygon by merging the intersecting circular buffers.

[0099] Specifically, to treat discrete outliers as structured spatial action zones, all edge outliers are extracted. Using each edge outlier as the center, a small increment is added to half of a set spatial clustering distance threshold (e.g., CLUSTER_DIST_THRESHOLD = 50m). Construct a circular buffer with radius. Perform a geometric merge and hidden surface removal operation (unary_union) on all circular buffers in the global domain, thereby merging spatially adjacent outlier point polygonal features into several independent spatially connected component polygons.

[0100] S103.3: For outlier points that fall within the same spatially connected component polygon, the nearest neighbor distance algorithm is used to sort and connect them to generate continuous outlier line segments.

[0101] Specifically, by leveraging spatial topological inclusion relationships, outlier point sets falling within the polygons of each connected component are extracted in reverse. Within each subset, any point is selected as the starting point, and the point set is spatially sorted using a nearest neighbor distance traversal algorithm. The sorted point sets are then connected sequentially to reconstruct continuous outlier line segment vector elements with topological coherence (such as...). Figure 6 (As shown). This process transforms scattered detection points into linear distribution areas with clear directions and lengths, thus defining a precise domain for subsequent targeted repair.

[0102] As one possible implementation, S104 specifically includes:

[0103] S104.1: Smooth the initial edge-joining results to obtain smoothed base data, and construct the spatial distance field between continuous abnormal line segments and the initial edge-joining results; Smooth the initial edge-joining results to obtain smoothed base data;

[0104] Specifically, after filling any holes that may exist in the initial edge-joining raster data using nearest-neighbor interpolation, a Gaussian low-pass filtering algorithm is used to perform global smoothing on the initial edge-joining results, with the blurred pixel radius set to [value missing]. in For the set buffer distance, (at raster resolution), to obtain smooth base data without numerical faults. The spatial distance from each pixel to the continuous abnormal line segment in the initial edge-connection result is calculated based on the Euclidean distance transformation algorithm. This yields a seamless spatial distance field.

[0105] S104.2: Generate different spatial adaptive decay fusion weights based on the spatial distance field, and use the different spatial adaptive fusion weights to perform weighted fusion on the initial edge repair results and the smooth basis function to obtain multiple sets of edge repair data.

[0106] Specifically, this invention provides four spatially adaptive decay fusion weights. These different weights are calculated using traditional Gaussian decay mode, cubic smooth step mode, quintic smooth step mode, and linear decay mode, respectively. Based on the configuration parameter `heal_method` input from the engineering project, and using the spatial distance field... Dynamically calculate the pixel-by-pixel attenuation fusion coefficient ( Between 0.0 and 1.0, at the core line of the seam. Outside the edge of the seam transition zone .

[0107] The spatial adaptive decay fusion weights corresponding to the traditional Gaussian decay mode The calculation formula is:

[0108]

[0109] in, This represents the distance from a pixel to a continuous abnormal line segment. Indicates the attenuation control parameter. This indicates the set buffer distance.

[0110] Spatial adaptive decay fusion weights corresponding to three smooth step modes The calculation formula is:

[0111]

[0112] In the formula,

[0113]

[0114] Specifically, a cubic polynomial smooth step function is used to eliminate the long-tail effect and achieve rapid boundary convergence.

[0115] Spatial adaptive decay fusion weights corresponding to five smooth step modes The calculation formula is:

[0116]

[0117] Specifically, the five-order smooth step mode introduces higher-order continuity.

[0118] Spatial adaptive decay fusion weights corresponding to linear decay modes The calculation formula is:

[0119] .

[0120] Extracting the initial edge-joining result raster matrix pixel by pixel With the corresponding smooth base raster matrix With adaptive decay weighting coefficient As spatial fusion coefficients, a pixel-wise linear weighted combination is performed to obtain the target raster data matrix after seamless restoration. :

[0121]

[0122] After blending, the original valid data mask is used to restore the original legitimate NoData region. This mechanism highly trusts the smoothed base data in the core contaminated area composed of continuous anomalous line segments to strongly heal the fault jump model, while preserving the real environmental texture of the original independent modeling results in the hinterland region far from the boundary line.

[0123] S104.3: Using the edge repair data of each group as the initial edge repair result, re-execute S102 and S103 respectively to generate the corresponding continuous abnormal line segments, and determine the final target raster data based on the number of continuous abnormal line segments in each group, specifically including:

[0124] Using the edge repair data of each group as the initial edge connection result, S102 and S103 are re-executed respectively to generate the corresponding continuous abnormal line segments as residual abnormal line segments, and the total length of each group of residual abnormal line segments is calculated.

[0125] Specifically, even after the smoothing process, the absolute value of the grade difference still exists in the statistics. The re-examination center points were identified, and the vector elements of "residual first-level abnormal line segments," "residual second-level abnormal line segments," and "residual severe abnormal line segments" corresponding to each mode were re-clustered and reconstructed. The total length of the residual abnormal line segments after repair for each mode was accurately calculated. .

[0126] Using formula Calculate the abnormal line segment elimination rate for each group, where This indicates the rate of elimination of abnormal line segments. This represents the total length of the continuous abnormal line segments before repair. This represents the total length of the remaining abnormal line segments; the edge repair data with the highest abnormal line segment elimination rate is selected as the target raster data.

[0127] Specifically, the abnormal line segment elimination rate of the calculation group is calculated, and the accuracy indicators such as the number of residual line segments and the total residual length under each mode are dynamically stored in a structured data table. After batch processing, the data table is formatted, and a comprehensive report of full-modal comparison named "Multi-algorithm Smoothing Repair Quality Comparison Summary.xlsx" is automatically exported. At the same time, when outputting Shapefiles (such as center detection points, sampling detail points) and repaired target raster TIF files to the hard drive, the corresponding algorithm name is automatically appended as an uppercase suffix to the file name. Finally, the horizontal evaluation table is retrieved, and the target raster data with the highest comprehensive abnormality elimination rate is automatically determined as the optimal edge-joining result for final mapping delivery (e.g., ...). Figure 7 (As shown).

[0128] Below is an example data table (as shown in Table 1) of the final horizontal comparison report of the repair quality of the full-modal algorithm, automatically generated in a large-scale soil property boundary engineering project:

[0129] Table 1. Examples of cross-modal comparison reports on the quality of multi-algorithm smooth repair.

[0130]

[0131] like Figure 8 As shown, this embodiment of the invention provides a cross-region modeling and edge-joining processing system based on mask constraints and multimodal adaptive attenuation fusion, comprising:

[0132] The initial edge-joining unit is used to acquire raster data generated independently from multiple target areas, unify the coordinate system of the raster data, and complete spatial stitching to obtain the initial edge-joining result.

[0133] The spatial sampling unit is used to extract the shared boundary between adjacent target regions from the initial edge-matching results, and to perform spatial sampling on both sides of the shared boundary to calculate the region cell value.

[0134] The anomaly identification unit is used to perform level mapping on the regional pixel values ​​according to the preset hierarchical system, identify the edge anomalies in the initial edge connection results according to the level mapping results, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering results.

[0135] The edge repair unit is used to perform spatial adaptive fusion repair on the initial edge repair results using continuous abnormal line segments to obtain the target raster data.

[0136] Those skilled in the art will understand that the specific parameters involved in the above embodiments (such as sampling step size, buffer distance, classification threshold, etc.) can be adjusted accordingly based on the spatial resolution of different geospatial data and the needs of specific business application scenarios, and do not constitute a limitation on the scope of protection of this invention.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion, characterized in that, include: Step 1: Obtain raster data generated independently from multiple target regions, unify the coordinate system of the raster data and complete spatial stitching to obtain the initial edge-joining result; Step 2: Extract the shared boundary between adjacent target regions from the initial edge-matching results, and perform spatial sampling on both sides of the shared boundary to calculate the region cell value; Step 3: Map the pixel values ​​of the region according to the preset hierarchical system, identify the edge anomalies of the initial edge connection result according to the hierarchical mapping result, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering result. Step 4: Using the continuous abnormal line segments, perform spatial adaptive fusion repair on the initial edge-joining results to obtain the target raster data.

2. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion as described in claim 1, characterized in that, Step 1 specifically includes: Acquire raster data and corresponding administrative division vector maps generated independently from multiple target areas, and unify the raster data of multiple target areas to the reference coordinate system through reprojection technology; The administrative vector boundaries of multiple target regions passing through coordinate system one are extracted, and the border area is determined outside the administrative vector boundary according to a preset distance. The border area of ​​multiple target regions is fused pixel by pixel using an overlapping filtering mean strategy to obtain the initial border result.

3. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 1, characterized in that, Step 1 specifically includes: Acquire raster data and corresponding administrative division vector maps generated independently from multiple target areas, and unify the raster data of multiple target areas to the reference coordinate system through reprojection technology; Extract the administrative vector boundaries of multiple target areas passing through coordinate system one, and crop and stitch the raster data of the target areas according to the administrative vector boundaries; The administrative division vector regions corresponding to the stitched raster data are converted into mask control matrices. Using the mask control matrix as a spatial constraint, an interpolation method based on inverse distance weights is employed to fill in the stitched raster data, resulting in the initial edge-joining result.

4. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 1, characterized in that, In step 2, spatial sampling is performed on both sides of the shared boundary to calculate the region cell value, including: On the shared boundary, sampling center points are extracted at preset intervals, and based on the orthogonal normal direction of the sampling center points on the shared boundary, they are extended to both sides of the region to a preset buffer distance to obtain normal sampling lines; Multiple sampling points are extracted along the normal sampling line with a preset fixed step size. The effective pixel statistical mean of each sampling point within its local neighborhood window is calculated. The arithmetic mean of the effective pixel statistical mean of all sampling points in a single-sided region of the normal sampling line is calculated as the region pixel value.

5. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 1, characterized in that, Step 3 specifically includes: Step 3.1: Map the pixel values ​​of the regions on both sides of the shared boundary to the corresponding discrete levels according to the preset hierarchical system, calculate the absolute value of the difference between the discrete levels on both sides, and determine the edge connection anomalies of the initial edge connection result based on the absolute value. Step 3.2: Construct a circular buffer with each of the aforementioned edge anomaly points as the center and a preset clustering topological distance as the radius, and obtain a spatially connected component polygon by merging the intersecting circular buffers; Step 3.3: For the edge anomalies that fall into the same spatially connected component polygon, sort and connect them using the nearest neighbor distance algorithm to generate continuous anomaly line segments.

6. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 1, characterized in that, Step 4 specifically includes: Step 4.1: Smooth the initial edge-joining result to obtain smoothed base data, and construct the spatial distance field between the continuous abnormal line segments and the initial edge-joining result; Step 4.2: Generate different spatial adaptive decay fusion weights based on the spatial distance field, and use the different spatial adaptive fusion weights to perform weighted fusion on the initial edge repair results and the smooth basis function to obtain multiple sets of edge repair data; Step 4.3: Using the edge repair data of each group as the initial edge repair results, re-execute steps 2 and 3 respectively to generate the corresponding continuous abnormal line segments, and determine the final target raster data based on the number of continuous abnormal line segments in each group.

7. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 6, characterized in that, Step 4.1 specifically includes: The initial edge-joining results are globally smoothed using a Gaussian low-pass filtering algorithm to obtain smoothed base data without numerical faults. The spatial distance field is obtained by calculating the spatial distance from each pixel in the initial edge-connection result to the continuous abnormal line segment based on the Euclidean distance transformation algorithm.

8. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 6, characterized in that, The different spatial adaptive decay fusion weights are calculated using the traditional Gaussian decay mode, the cubic smooth step mode, the quintic smooth step mode, and the linear decay mode, respectively; wherein the spatial adaptive decay fusion weights corresponding to the traditional Gaussian decay mode are... The calculation formula is: in, This represents the distance from a pixel to a continuous abnormal line segment. Indicates the attenuation control parameter. This indicates the set buffer distance; The spatial adaptive decay fusion weights corresponding to the three smooth step modes The calculation formula is: In the formula, ； The spatial adaptive decay fusion weights corresponding to the five smooth step modes The calculation formula is: ； The spatial adaptive decay fusion weight corresponding to the linear decay mode The calculation formula is: 。 9. The cross-region modeling and edge-joining method based on mask constraint and multimodal adaptive attenuation fusion according to claim 6, characterized in that, Step 4.3 specifically includes: Using the edge repair data of each group as the initial edge repair results, steps 2 and 3 are re-executed to generate corresponding continuous abnormal line segments as residual abnormal line segments. The total length of each group of residual abnormal line segments is calculated, and the formula is used to... Calculate the abnormal line segment elimination rate for each group, where This indicates the rate of elimination of abnormal line segments. This represents the total length of the continuous abnormal line segments before repair. Indicates the total length of the remaining abnormal line segments; The edge repair data with the highest abnormal line segment elimination rate was selected as the target raster data.

10. A cross-region modeling and edge-joining system based on mask constraint and multimodal adaptive attenuation fusion, characterized in that, include: The initial edge-joining unit is used to acquire raster data generated independently by modeling multiple target areas, unify the coordinate system of the raster data and complete spatial stitching to obtain the initial edge-joining result; A spatial sampling unit is used to extract the shared boundary between adjacent target regions from the initial edge-connection result, and to perform spatial sampling on both sides of the shared boundary to calculate the region cell value; An anomaly identification unit is used to perform level mapping on the pixel values ​​of the region according to a preset hierarchical system, identify the edge anomalies of the initial edge connection result according to the level mapping result, perform spatial clustering on the edge anomalies, and generate continuous abnormal line segments based on the spatial clustering result. The edge repair unit is used to perform spatial adaptive fusion repair on the initial edge repair result using the continuous abnormal line segments to obtain the target raster data.

Images